Research On Multi-type Vehicle Detection And DIC High-precision Vehicle Tracking Based On Monocular Vision

Due to the frequent occurrence of traffic accidents in China,people's attention to driving safety has been further improved.Therefore,the forward collision warning system has become a research hotspot.The accuracy of vehicle detection,accuracy of tracking and real-time performance are important evaluation indicators of the system.Therefore,this paper focuses on the front-vehicle detection and high-precision tracking algorithm based on monocular vision,which provides more guarantee for driving safety.The main research work is as follows:1.A method for multi-classifier series connection is proposed to solve the problem that the single Adaboost cascade classifier,the detection time is long when the number of stages is high,and the robustness of simultaneous detection of multiple types of vehicles(cars,trucks,buses,etc)is not high.Compared with a single cascade classifier,the robustness of the method in multi-type vehicle detection under weather conditions such as sunny,cloudy and light rain is verified by experiments.2.A high-precision vehicle tracking algorithm based on digital image correlation(DIC)is proposed.The tracking accuracy can reach 0.001 pixels to solve the problem that the tracking of the preceding vehicle is not accurate and the collision warning is not timely.By comparing with the CMT tracking algorithm and adaptive Mean Shift algorithm in tracking frame width,coordinates,tracking time and so on,the accuracy of the algorithm tracking on the vehicle is verified.The tracking time per frame is only 0.995 ms.3.Design a set of pre-collision warning system to verify the feasibility of the algorithm.By setting different relative speeds of the two cars,the alarm distance is counted,and the timeliness of the system early warning is verified,and the real-time and effectiveness of the algorithm are verified.